Wetland to Upland Gradient
Anthony Stewart
Questions:
Does a model derived gradient between wetlands and uplands correspond to a gradient of soil development along a catena/hillslope characterized in a hydropedological framework?
If the ends of the wetland/upland classification are accurate, does the intermediate range correspond with hypothesized soil formation characteristics according to a hydropedological framework?
hydropedology hillslope model from Bailey et
al., 2014
The Hubbard Brook Experimental Forest
From Hubbard Brook website https://hubbardbrook.org/about-the-forest/:
- “The Hubbard Brook Experimental Forest is located on unceded Abenaki land in the southwestern part of the White Mountain National Forest in New Hampshire. The experimental forest occupies a 3,037-hectare (~8,700-acre) bowl-shaped valley that is oriented east–west, with a series of small watersheds occupying the valley’s north- and south-facing slopes.”
Hydropedological units in the Hubbard Brook Experimental Forest include:
- Drier more upland soils: Bhs, Typical (Podzol)
- Wetter soils: Bh, Histosol
- Intermediate/mesic soils: Bimodal, Inceptisol
Gather and filter NWI data
## [1] "ATTRIBUTE" "WETLAND_TY" "ACRES" "Shape_Leng" "Shape_Area"
## SpatRaster object downsampled to 775 by 1291 cells.
Gather Hubbard Brook hydrography
- Hydrography is lines, so we need to buffer to make them polygons. We use 1m buffer
- To reduce overlap we remove Riverine wetlands from the NWI source
Get training data - wetland and upland points
Total area is 5816.7781 ha Wetland area for training data is 66.63021 ha (~10%)
Non-wetland areas or “upland” areas have removed wetland/wet areas with a 5m buffer to reduce potential mis-classification. Some wet areas are not identified yet, which means some potential non-wetland areas may contain wet areas. Making the non-wetland point dataset larger can offset this a bit
## [1] 1947
## SpatRaster object downsampled to 775 by 1291 cells.
## [1] 530
## SpatRaster object downsampled to 775 by 1291 cells.
Make terrain metrics
hbslp_3 <- SlpAsp(hbdem, w = c(3, 3), metrics = "slope", filename = "UplandWetlandGradient/data/derived_data/hbslp_3.tif",
overwrite = T)
hbslp_27 <- SlpAsp(hbdem, w = c(27, 27), metrics = "slope", filename = "UplandWetlandGradient/data/derived_data/hbslp_27.tif",
overwrite = T)
hbslp_81 <- SlpAsp(hbdem, w = c(81, 81), metrics = "slope", filename = "UplandWetlandGradient/data/derived_data/hbslp_81.tif",
overwrite = T)
hbtpi_3 <- TPI(hbdem, w = c(3, 3), shape = "rectangle", stand = "none",
na.rm = TRUE, filename = "UplandWetlandGradient/data/derived_data/hbtpi_3.tif",
overwrite = T)
hbtpi_27 <- TPI(hbdem, w = c(27, 27), shape = "rectangle", stand = "none",
na.rm = TRUE, filename = "UplandWetlandGradient/data/derived_data/hbtpi_27.tif",
overwrite = T)
hbtpi_81 <- TPI(hbdem, w = c(81, 81), shape = "rectangle", stand = "none",
na.rm = TRUE, filename = "UplandWetlandGradient/data/derived_data/hbtpi_81.tif",
overwrite = T)
hbcurv_3 <- Qfit(hbdem, w = c(3, 3), metrics = c("meanc", "profc",
"planc"), unit = "degrees", na.rm = TRUE, filename = "UplandWetlandGradient/data/derived_data/hbcurv_3.tif",
overwrite = T)
hbcurv_27 <- Qfit(hbdem, w = c(27, 27), metrics = c("meanc",
"profc", "planc"), unit = "degrees", na.rm = TRUE, filename = "UplandWetlandGradient/data/derived_data/hbcurv_27.tif",
overwrite = T)
hbcurv_81 <- Qfit(hbdem, w = c(81, 81), metrics = c("meanc",
"profc", "planc"), unit = "degrees", na.rm = TRUE, filename = "UplandWetlandGradient/data/derived_data/hbcurv_81.tif",
overwrite = T)hbasp_3 <- Qfit(hbdem, w = c(3, 3), metrics = c("qaspect"), unit = "degrees",
na.rm = TRUE, filename = "UplandWetlandGradient/data/derived_data/hbasp_3.tif",
overwrite = T)
hbhli_3 <- spatialEco::hli(hbasp_3)
hbasp_27 <- Qfit(hbdem, w = c(27, 27), metrics = c("qaspect"),
unit = "degrees", na.rm = TRUE, filename = "UplandWetlandGradient/data/derived_data/hbasp_27.tif",
overwrite = T)
hbhli_27 <- spatialEco::hli(hbasp_27)
hbasp_81 <- Qfit(hbdem, w = c(81, 81), metrics = c("qaspect"),
unit = "degrees", na.rm = TRUE, filename = "UplandWetlandGradient/data/derived_data/hbasp_81.tif",
overwrite = T)
hbhli_81 <- spatialEco::hli(hbasp_81)Slope at different scales
TPI at different scales
Curvature at different scales
Aspect/HLI at different scales
Whitebox tools Deviation from mean elevation
Is resampling the same as setting the window size larger?
## [1] "class" "dem1m" "meanc" "profc" "planc"
## class : SpatVector
## geometry : points
## dimensions : 1495, 5 (geometries, attributes)
## extent : 275091.5, 283296.5, 4866362, 4871276 (xmin, xmax, ymin, ymax)
## coord. ref. : NAD83 / UTM zone 19N (EPSG:26919)
## names : class dem1m meanc profc planc
## type : <chr> <num> <num> <num> <num>
## values : UPL 688.7 -0.001151 -0.0001594 -0.002142
## UPL 584 0.001018 0.001373 0.0006639
## UPL 581.4 -0.0007899 -0.001685 0.0001053
Hydrology metrics
Hydrologic conditioning
D-infinity flow accumulation
Topographic Wetness Index
- with different scale slopes?
Depth to water from cost distance function
- DTW is too biased of a predictor for streams because it creates an almost linear metric related to stream location.
- Doing this with including NWI wetlands further biases towards training data.
Stack rasters and use point data for extraction
Set up random forest model
## randomForest 4.7-1.1
## Type rfNews() to see new features/changes/bug fixes.
##
## Attaching package: 'randomForest'
## The following object is masked from 'package:dplyr':
##
## combine
## The following object is masked from 'package:spatialEco':
##
## combine## Loading required package: ggplot2
## Warning: package 'ggplot2' was built under R version 4.3.1
##
## Attaching package: 'ggplot2'
## The following object is masked from 'package:randomForest':
##
## margin
## Loading required package: lattice
## Confusion Matrix and Statistics
##
## Reference
## Prediction UPL WET
## UPL 711 33
## WET 17 326
##
## Accuracy : 0.954
## 95% CI : (0.9398, 0.9657)
## No Information Rate : 0.6697
## P-Value [Acc > NIR] : < 2e-16
##
## Kappa : 0.8948
##
## Mcnemar's Test P-Value : 0.03389
##
## Sensitivity : 0.9766
## Specificity : 0.9081
## Pos Pred Value : 0.9556
## Neg Pred Value : 0.9504
## Prevalence : 0.6697
## Detection Rate : 0.6541
## Detection Prevalence : 0.6845
## Balanced Accuracy : 0.9424
##
## 'Positive' Class : UPL
##
## Confusion Matrix and Statistics
##
## Reference
## Prediction UPL WET
## UPL 274 24
## WET 24 145
##
## Accuracy : 0.8972
## 95% CI : (0.866, 0.9232)
## No Information Rate : 0.6381
## P-Value [Acc > NIR] : <2e-16
##
## Kappa : 0.7775
##
## Mcnemar's Test P-Value : 1
##
## Sensitivity : 0.9195
## Specificity : 0.8580
## Pos Pred Value : 0.9195
## Neg Pred Value : 0.8580
## Prevalence : 0.6381
## Detection Rate : 0.5867
## Detection Prevalence : 0.6381
## Balanced Accuracy : 0.8887
##
## 'Positive' Class : UPL
##
## [1] 29.776
AUC/ROC
## Type 'citation("pROC")' for a citation.
##
## Attaching package: 'pROC'
## The following objects are masked from 'package:stats':
##
## cov, smooth, var
##
## Attaching package: 'ROCR'
## The following object is masked from 'package:performance':
##
## performance
## [1] "Area under the ROC curve"
## [[1]]
## [1] 0.8887256
## [1] 0.8972163
## [1] 0.9194631
## Warning in confusionMatrix.default(as.factor(test_df$wetupl),
## as.factor(test_df$test_class)): Levels are not in the same order for reference
## and data. Refactoring data to match.## `geom_smooth()` using method = 'loess' and formula = 'y ~ x'
PDF/kernel density plot
Boosted Regression Trees
## var rel_inf
## dev81 dev81 56.56816162
## slp_27 slp_27 11.98661984
## slp_81 slp_81 5.78882225
## dem dem 5.41586322
## dev27 dev27 4.37506444
## plan_curv_27 plan_curv_27 2.91163285
## meancurv_81 meancurv_81 2.72417947
## plan_curv_81 plan_curv_81 2.21375416
## tpi_27 tpi_27 1.76460306
## slp_3 slp_3 1.10757720
## plan_curv_3 plan_curv_3 0.92665550
## prof_curv_81 prof_curv_81 0.91851155
## tpi_81 tpi_81 0.77612130
## prof_curv_27 prof_curv_27 0.72856500
## meancurv_27 meancurv_27 0.31287102
## hli_81 hli_81 0.29643587
## hb_sca_log hb_sca_log 0.28151554
## hli_3 hli_3 0.22607859
## twi twi 0.22502912
## hli_27 hli_27 0.10857315
## tpi_3 tpi_3 0.10689530
## meancurv_3 meancurv_3 0.09895650
## prof_curv_3 prof_curv_3 0.08669080
## dev3 dev3 0.05082266
## Confusion Matrix and Statistics
##
## Reference
## Prediction UPL WET
## UPL 269 24
## WET 29 145
##
## Accuracy : 0.8865
## 95% CI : (0.8542, 0.9138)
## No Information Rate : 0.6381
## P-Value [Acc > NIR] : <2e-16
##
## Kappa : 0.7558
##
## Mcnemar's Test P-Value : 0.5827
##
## Sensitivity : 0.9027
## Specificity : 0.8580
## Pos Pred Value : 0.9181
## Neg Pred Value : 0.8333
## Prevalence : 0.6381
## Detection Rate : 0.5760
## Detection Prevalence : 0.6274
## Balanced Accuracy : 0.8803
##
## 'Positive' Class : UPL
## 
Predict maps
## SpatRaster object downsampled to 775 by 1291 cells.
## SpatRaster object downsampled to 775 by 1291 cells.
## # A tibble: 3 × 2
## kmeanclus meanWET
## <fct> <dbl>
## 1 1 0.544
## 2 2 0.266
## 3 3 0.119
Histogram of raster
## Warning: [hist] a sample of 17% of the cells was used (of which 47% was NA)
